A method for overcoming signal interference in indoor positioning in home medical care

Abstract

The application discloses a kind of indoor positioning method for overcoming signal interference in home medical care, belong to indoor positioning technical field.The method is by adding fuzzy space with weakening boundary absolute difference characteristics, establish the update index of signal pair space discrimination influence, to avoid the belonging misjudgment phenomenon of the similarity difference of cluster is not significant TP, improve the efficiency and precision of positioning, to meet the requirement of positioning accuracy and timeliness in home medical care;When the matching of to-be-positioned point and reference point is carried out, the similarity of the two is no longer measured by Euclidean distance, but by combining the positioning information carried by RP neighborhood space, weakening the influence of signal fluctuation, and then realizing the judgment of multi-source balanced similarity between TP and RP by robust ordering of homologous similarity of RP.And a positioning method of signal domain and space domain iterative constraint neighbor point set is also proposed, which improves the selection accuracy of neighbor RP while ensuring its relatively concentrated distribution, thereby ensuring the effectiveness of positioning.

Description

Technical Field

[0001] This invention relates to an indoor positioning method for overcoming signal interference in home medical care, belonging to the field of indoor positioning technology. Background Technology

[0002] With economic and social development and the improvement of human engineering capabilities, the importance of location-based services has permeated all aspects of society. Especially in home healthcare monitoring systems, accurately locating the user's home coordinates is essential and crucial. Proactive medical assistance, spontaneous monitoring and tracking, and smart home solutions for the elderly all require known user or target location information as a prerequisite. However, existing positioning technologies cannot meet the accuracy requirements of home healthcare monitoring systems.

[0003] As is well known, the continuous maturation of Global Navigation Satellite Systems (GNSS) has greatly supported a wide range of positioning applications in outdoor scenarios. However, due to satellite signal obstruction caused by buildings, the satellite positioning technology commonly used outdoors cannot be easily transferred to complex indoor positioning scenarios, especially in applications like home medical monitoring systems where accuracy, stability, and timeliness are critical. Therefore, low-error, high-reliability indoor positioning technology has become a key research focus for home medical monitoring systems. With the widespread deployment of indoor WiFi sources and the popular application of smart mobile devices, fingerprint positioning algorithms based on signal strength have become one of the most commonly used methods due to their high cost-effectiveness and environmental universality. This method, in the offline stage, completes the marking of the locations of each Reference Point (RP) and the sample collection and analysis of the Received Signal Strength (RSS) of its visible Access Points (APs) to construct a fingerprint database. In the online phase, the RSS vectors received in real time from the points to be located (or test points, TP) are matched with samples in the fingerprint database built in the offline phase to achieve spatial location estimation of the target.

[0004] Traditional fingerprint localization methods typically involve sequentially matching the fingerprint probe (TP) with the fingerprint reference points (RPs) in the fingerprint database based on similarity. This approach incurs high computational costs and significantly impacts localization accuracy due to the distribution of RPs. To reduce computational burden and improve accuracy, clustering algorithms have been introduced into fingerprint localization. In the offline phase, the fingerprint database is clustered using methods such as K-means, fuzzy C-means (FCM), and affinity propagation clustering (APC) to divide the entire target area into multiple sub-regions. This clustering approach enables online cluster matching, replacing the global comparison process with similarity matching between the TP and the RPs within its own cluster. However, unlike simple data clustering, the wireless channel, as the data transmission channel, exhibits unpredictable dynamic attenuation and multipath effects. This makes it difficult for fingerprint information to have absolutely clear inter-cluster boundaries, and the inter-cluster transition zone significantly influences the misclassification of edge point information.

[0005] Besides offline clustering factors, online matching plays a crucial role in the effectiveness of localization. In the offline phase, the average received signal strength from multiple acquisitions at a single reference point (RP) is typically calculated to overcome the problem of dynamic signal changes. However, in actual online localization, users usually only stay at a point for 1-2 seconds, so the uploaded data may not accurately reflect the true RSS value of the current location due to noise. The Weighted k-nearest neighbor (WKNN) algorithm estimates the location by calculating the weighted centroid of the K nearest neighbor RPs with the highest similarity in the fingerprint database. However, it lacks consideration for signal fluctuations caused by environmental characteristics such as target movement, obstacle changes, and AP state changes. Furthermore, a fixed K value and dispersed fingerprint point selection cannot avoid the loss of important spatial information or even the introduction of false outliers. To adapt to dynamic environmental changes, multi-objective optimization is used to construct a criterion model to achieve adaptive selection of online APs; or reference point ranking is used to mitigate the instability of the similarity judgment standard. However, these solutions lack consideration for the impact of dynamic signal differences on localization. To address the shortcomings of fixed reference point selection in adapting to dynamic changes in fingerprints, a dynamic K-value concept is introduced to remove fingerprint points with Euclidean similarity less than a threshold to ensure a certain level of positioning accuracy. However, due to signal uncertainty, it is still impossible to ensure the removal of outliers. Alternatively, the concentration of nearest neighbor locations can be enhanced by processing the received signal strength from each access point individually, but the algorithm has high time complexity and real-time performance cannot be guaranteed. Summary of the Invention

[0006] To address the existing problems of blurred inter-cluster boundaries and cross-interference of fingerprint positioning results by environmental and signal dynamic characteristics, this invention provides an indoor positioning method to overcome signal interference in home healthcare, the method comprising an offline stage and an online stage;

[0007] In the offline phase, the fingerprint database is constructed, and the reference points in the fingerprint database are initially clustered using the average signal strength as a single scale to generate T clusters, and the corresponding cluster centers are recorded; the reference points located at the cluster edges are then clustered a second time using the "inter-cluster-intra-cluster" dual distance.

[0008] In the online phase, the RSS vector received in real time by the point to be located (TP) is matched with the samples in the fingerprint database constructed in the offline phase to estimate the spatial location of the target. During the matching process, the similarity between the point to be located and the fingerprint points is measured by the Euclidean distance between the points. At the same time, the positioning information carried by the nearest neighbor space of the RP is combined to weaken the influence of signal fluctuations. The RP is robustly ranked according to the similarity of the same source to achieve accurate estimation of the spatial location of the target.

[0009] Optionally, the method includes:

[0010] Step S1: Construct the system model;

[0011] Offline Phase: Assume that there are N reference points (RPs) and M access points (APs) deployed in the planar space where the point to be located is located; RP n The physical location is denoted as (x n ,y n ), where n∈{1,2,...,N}, and in each RP n The above samples Q signal strength values ​​from M non-line-of-sight access points (APs), denoted as .

[0012] Then in RP n The location received information from the source AP. m The average signal strength is With RP n The average signal strength characterizes the location features of this point, represented as a vector. And record fingerprint information The fingerprint database contains fingerprint information for all reference points (RPs).

[0013] Online Phase: The signal strength sample values ​​received by the target point TP at (x', y') from M access points AP are denoted as RSS′=[rss'1, rss'2, ..., rss' m ,...,rss' M ];

[0014] Step S2, spatial fuzzy partitioning;

[0015] Initial clustering is performed using the average signal intensity at each reference point as a single scale, generating T clusters Ω1,Ω2,...,Ω T And record its cluster centers c1, c2, ..., cT Reference points belonging to the same cluster are called a class; the planar space is divided into T sub-regions Ω1, Ω2, ..., Ω according to their class. T ;

[0016] Calculate the inter-class distance b for each reference point. (t) (s,i) and intra-class distance w(s,i); b (t) (s,i) represents the i-th fingerprint point in the s-th class. up to all n in class t t The average signal domain distance of the fingerprint points; s and t represent class labels, s,t∈{1,2,...,T}, s≠t; w(s,i) represents the i-th fingerprint point in the s-th class. To the other n in this class s -1 average signal domain distance of fingerprint points;

[0017] Set a spatial ambiguity threshold f, based on the inter-class distance b between each reference point. (t) The relationship between the absolute value of the difference between (s,i) and the intra-class distance w(s,i) and the spatial fuzziness threshold f determines the reference point. Should it be added to subregion Ω? t Internally, implement category updates;

[0018] Step S3: Construct a dual-label fingerprint database;

[0019] The initial clusters Ω1, Ω2, ..., Ω are generated during the initial clustering. T As the class label for the target region, the updated RP set contained in each class is used as the in-class point label, and points belonging to Ω are assigned to it. t The inner point label of the class is denoted as Where G represents Ω t The updated number of in-class point labels;

[0020] A dual-label offline fingerprint database is constructed by combining class labels, intra-class point labels, and their fingerprint information.

[0021] Step S4: Position calculation, to obtain the discrimination sub-region and estimated position of the point TP to be located;

[0022] The signal strength values ​​from each AP are collected at the point to be located (TP), and the nearest sub-cluster (Ω) is matched to TP based on the collected signal strength values. t As the home region, Ω t The RP set contained within As candidate fingerprint points;

[0023] For those already assigned to Ω t TP, calculate its relationship with Ω t middle The absolute difference in received signal strength is used to map the similarity of origin.

[0024] For the same source AP m , TP and Ω t All RP pairs within AP m The similarity set is denoted as This is used to obtain the fingerprint points corresponding to the elements of the set in descending order. At the source AP m Relative sorting values

[0025] To systematically evaluate the ranking value of origin similarity, all access points are paired with reference points. The ranking effect is considered in multi-source equilibrium. As The final sorting;

[0026] Introducing spatial density reachability discrimination, searching for strongly correlated reference sets (RPs), and adaptively controlling the K value, we determine the effective reference set for the point TP to be located. Then, estimate the planar spatial coordinates (x', y') of the point TP to be located:

[0027]

[0028] in, Indicates the effective nearest neighbor. The sorting value.

[0029] Optionally, when performing spatial fuzzy partitioning in step S2, the inter-class distance b (t) The formula for calculating (s,i) is:

[0030]

[0031] in, and They represent and The received signal sample value;

[0032] The formula for calculating the intra-class distance w(s,i) is:

[0033]

[0034] in, express The received signal sample value, and j≠i.

[0035] Optionally, the step of basing the data on the inter-class distance b between each reference point... (t)The relationship between the absolute value of the difference between (s,i) and the intra-class distance w(s,i) and the spatial fuzziness threshold f determines the reference point. Should it be added to subregion Ω? t Internally, category updates are implemented, including:

[0036]

[0037] Optionally, in step S4, the homology similarity... The calculation formula is:

[0038]

[0039] Among them, the additional term σ>0 and σ→0.

[0040] Optionally, in step S4, the fingerprint points corresponding to the elements of the set are obtained in descending order. At the source AP m Relative sorting values Then, it also includes: sorting the samples from the same source neighbor in ascending order and then performing quartile processing, denoted as Q. u and Q d The upper and lower quartiles are the interquartile range (IQR) = Q. u -Q d Based on Q d and Q u Scaling the IQR by 1.5 times to divide the effective interval, constructing the home region Ω. t Outlier correction model for internal RP homology sorting:

[0041]

[0042] In the formula, express Non-abnormal sorted neighbor points;

[0043] like Located outside the boundary, update it to The average of the sorted values; otherwise, Keep the original value.

[0044] Optionally, in step S4, a spatial density reachability determination is introduced, a strongly correlated reference set (RP) is searched, the K value is controlled to adapt, and an effective reference set for the point TP to be located is determined. include:

[0045] 1) Search for strongly correlated RP sets

[0046] Get Home Zone Ω t From the sorting results of all RP and TP, take the first I reference points to generate the initial screening nearest neighbor set RP' = {RP'1, RP'2, ..., RP'}. i,...,RP' I};

[0047] The spatial domain distance of the initial nearest neighbor points is used to apply secondary constraints to the target nearest neighbor set;

[0048] To screen nearest neighbor RP' i Define RP' with the spatial coordinates as the center and ε as the radius. i The neighborhood is:

[0049] Nei ε (RP' i )={RP' j ∈RP'|dist(RP' i ,RP' j )≤ε} (7)

[0050] Among them, dist(RP') i ,RP' j ) represents two initially screened nearest neighbors RP' in set RP'. i and RP' j Spatial domain distance between them, Nei ε (RP' i ) contains RP' and RP' i All initially screened nearest neighbors whose spatial distance is not greater than ε;

[0051] Set a threshold η in the domain to reflect RP' i The density of surrounding neighboring points, if |Nei ε (RP' i If |≥η, then RP' i Considered a core point (CP); if a core point sequence exists. Then, according to equation (8), it is determined whether the preceding term of an adjacent core point in the sequence is within the neighborhood of the following term, and labeled with l. h Logo:

[0052]

[0053] The resulting H-1 labels collectively determine the core point sequence {CP1, CP2, ..., CP...} h ,...,CP H Whether a set of neighbors has the potential to form a target nearest neighbor set is defined as follows:

[0054]

[0055] If L = 1, then CP HIf the density of CP1 is reachable, it indicates that the spatial domain of the core point sequence is relatively close, and the fingerprint information it carries has a high degree of similarity, making it possible to form a target nearest neighbor set; otherwise, it does not meet the conditions for forming a set.

[0056] Starting from any core point CP, a cluster is formed from all CPs that the density of that CP can reach; then, new starting points that can be clustered are found from the remaining CPs that have not yet been clustered, thus completing the search for the candidate set of strongly correlated RPs. Points that cannot be clustered are regarded as weakly correlated reference points.

[0057] 2) Target nearest neighbor matching

[0058] Using the number of target CPs in a strongly correlated RP set, supplemented by their average signal domain distance to the target TP, a triple constraint is applied to the target nearest neighbor set to select the set of targets with the highest probability:

[0059] If a strongly correlated RP set has the smallest average signal domain distance to TP and contains the most CPs, then it is retained as the target nearest neighbor set. Where K represents the number of RPs in the target nearest neighbor set, 2≤K≤I; otherwise, the initial screening nearest neighbor set is output for final location estimation.

[0060] Optionally, if the initial screening nearest neighbor set of a certain TP is entirely judged as a weakly correlated reference point, then the initial screening nearest neighbor set is used as the target nearest neighbor set for the location estimation of that TP.

[0061] This application also provides the application of the above-mentioned indoor positioning method in rehabilitation and elderly care.

[0062] The beneficial effects of this invention are:

[0063] In the indoor positioning method provided in this application, considering the stringent positioning accuracy requirements of home medical monitoring systems, and addressing the issue that existing indoor positioning methods often use an either-or approach when dividing the entire target area, leading to the incorrect assignment of points at cluster boundaries to neighboring clusters and resulting in significant positioning errors that fail to meet the positioning accuracy requirements of home medical monitoring systems, this application adds a fuzzy space with weakened absolute boundary differences and establishes an update index for the signal's influence on spatial discrimination. This avoids misclassification of TPs with insignificant inter-cluster similarity differences, thereby improving positioning accuracy. Furthermore, in the matching process between the target point and the reference point, it no longer limits the measurement of similarity between the target point and the fingerprint point to Euclidean distance, but instead combines... By combining the positioning information carried by the nearest neighbor space of the RP (Real Point), the influence of signal fluctuations is weakened, and the RPs are robustly ranked according to their similarity. This enables effective judgment of the multi-source balanced similarity between the RPs and the TP (Target Point), overcoming the positioning deviation caused by signal interference. Furthermore, this application proposes a positioning method that iteratively constrains the nearest neighbor set in both the signal and spatial domains. By introducing spatial density reachability discrimination, it searches for strongly correlated RP sets and controls the K value to be adaptive, improving the accuracy of nearest neighbor RP selection while ensuring their relatively concentrated distribution. Compared with existing global greedy matching methods, this application's method greatly ensures the real-time positioning while meeting the positioning accuracy requirements of home medical monitoring systems, providing maximum protection for timely treatment of some emergencies in home medical monitoring systems. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1A This is a schematic diagram of planar hard clustering used in existing indoor positioning methods;

[0066] Figure 1B This is a schematic diagram of soft clustering in planar space proposed in the positioning method of this application.

[0067] Figure 2A This is a schematic diagram of the initial constraint process for point matching in the positioning method of this application;

[0068] Figure 2B This is a schematic diagram of the point matching quadratic constraint for the positioning method in this application;

[0069] Figure 2C The point matching method of this application is shown in a cubic constraint diagram.

[0070] Figure 3 This is a flowchart of the DLFS dual-label fingerprint localization method in fuzzy space provided in one embodiment of the present invention.

[0071] Figure 4 The figure shows the simulation results of the impact of the fuzzy threshold f on the region discrimination accuracy and the global RP growth rate in the positioning method proposed in this invention.

[0072] Figure 5A This is a heatmap of positioning error under different combinations of domain parameters (ε,η) within sub-region Ω2 in one embodiment of the present invention;

[0073] Figure 5B This is a heatmap of positioning error under different combinations of field parameters (ε,η) within sub-region Ω5 in one embodiment of the present invention;

[0074] Figure 5C This is a heatmap of positioning error under different combinations of field parameters (ε,η) within sub-region Ω7 in one embodiment of the present invention.

[0075] Figure 6 This is a schematic diagram of soft clustering results for a certain area within an indoor planar space in a home maintenance scenario provided in one embodiment of the present invention.

[0076] Figure 7A The simulation comparison chart shows the cumulative probability distribution of the positioning error of the test point obtained by indoor positioning using the method of the present invention and the global matching method, respectively.

[0077] Figure 7B The simulation comparison diagram shows the cumulative probability distribution of the positioning error of the test point obtained by indoor positioning using the method of the present invention and the local matching method, respectively. Detailed Implementation

[0078] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0079] In various indoor healthcare settings such as hospitals, nursing homes, communities, and homes, accurate indoor positioning can be used to create enhanced digital experiences in healthcare settings. For example, accurate indoor positioning is crucial for patient navigation, user tracking, healthcare worker tracking, device search, data analysis, and facility management. Knowing the location of the user or target allows staff to focus on their core tasks, improving efficiency; patients receive more timely and accurate assistance; and accurate positioning information also maximizes device utilization and optimizes scenario management.

[0080] Example 1:

[0081] This embodiment provides an indoor positioning method to overcome signal interference in home-based medical care, the method comprising:

[0082] Step S1: Construct the system model;

[0083] Offline Phase: Assume that there are N reference points (RPs) and M access points (APs) deployed in the planar space where the point to be located is located; RP n The physical location is denoted as (x n ,y n ), where n∈{1,2,...,N}, and in each RP n The above samples Q signal strength values ​​from M non-line-of-sight access points (APs), denoted as .

[0084] Then in RP n The location received information from the source AP. m The average signal strength is With RP n The average signal strength characterizes the location features of this point, represented as a vector. And record fingerprint information The fingerprint database contains fingerprint information for all reference points (RPs).

[0085] Online Phase: The signal strength sample values ​​received by the target point TP at (x', y') from M access points AP are denoted as RSS′=[rss'1, rss'2, ..., rss' m ,...,rss' M ];

[0086] Step S2, spatial fuzzy partitioning;

[0087] Initial clustering is performed using the average signal intensity at each reference point as a single scale, generating T clusters Ω1,Ω2,...,Ω T And record its cluster centers c1, c2, ..., c T Reference points belonging to the same cluster are called a class; the plane space to which they belong is divided into T sub-regions Ω1, Ω2, ..., Ω according to their class. T ;

[0088] Calculate the interclass distance b between reference points within the target area. (t) (s,i) and intra-class distance w(s,i); b (t) (s,i) represents the i-th fingerprint point in the s-th class. up to all n in class t t The average signal domain distance of the fingerprint points; s and t represent class labels, s,t∈{1,2,...,T}, s≠t; w(s,i) represents the i-th fingerprint point in the s-th class. To the other n in this class s-1 is the average signal domain distance of fingerprint points; fingerprint points are the reference points.

[0089] Set a spatial ambiguity threshold f, based on the inter-class distance b between each reference point. (t) The relationship between the absolute value of the difference between (s,i) and the intra-class distance w(s,i) and the spatial fuzziness threshold f determines the reference point. Should it be added to subregion Ω? t Internally, implement category updates;

[0090] Step S3: Construct a dual-label fingerprint database;

[0091] The initial clusters Ω1, Ω2, ..., Ω are generated during the initial clustering. T As the class label for the target region, the updated RP set contained in each class is used as the in-class point label, and points belonging to Ω are assigned to it. t The inner point label of the class is denoted as Where G represents Ω t The updated number of in-class point labels;

[0092] A dual-label offline fingerprint database is constructed by combining class labels, intra-class point labels, and their fingerprint information.

[0093] Step S4: Position calculation, to obtain the discrimination sub-region and estimated position of the point TP to be located;

[0094] The signal strength values ​​from each AP are collected at the point to be located (TP), and the nearest sub-cluster (Ω) is matched to TP based on the collected signal strength values. t As the home region, Ω t The RP set contained within As candidate fingerprint points;

[0095] For those already assigned to Ω t TP, calculate its relationship with Ω t middle The absolute difference in received signal strength is used to map the similarity of origin.

[0096] For the same source AP m , TP and Ω t All RP pairs within AP m The similarity set is denoted as This is used to obtain the fingerprint points corresponding to the elements of the set in descending order. At the source AP m Relative sorting values

[0097] To systematically evaluate the ranking value of origin similarity, all access points are paired with reference points. The ranking effect is considered in multi-source equilibrium. As The final sorting;

[0098] Introducing spatial density reachability discrimination, searching for strongly correlated reference sets (RPs), and adaptively controlling the K value, we determine the effective reference set for the point TP to be located. Then, estimate the planar spatial coordinates (x', y') of the point TP to be located:

[0099]

[0100] in, Indicates the effective nearest neighbor. The sorting value.

[0101] Example 2:

[0102] This embodiment provides an indoor positioning method to overcome signal interference in home-based medical care, the method comprising:

[0103] Step 1: System Model Construction;

[0104] In the offline phase, assuming N reference points RP are deployed in planar space, RP n The physical location is denoted as (x n ,y n ), where n∈{1,2,...,N}, and in each RP n The above samples Q signal strength values ​​from M non-line-of-sight access points (APs). Therefore, in RP n The location received information from the source AP. m The average signal strength is With RP n The average signal strength characterizes the location features of this point, represented as a vector. And record fingerprint information

[0105] During the online phase, the target point TP located at (x', y') receives signal strength sample values ​​RSS′=[rss'1, rss'2, ..., rss' m ,...,rss' M ].

[0106] Step 2: Spatial fuzzy partitioning;

[0107] This application proposes the concept of fuzzy partitioning when performing spatial division. Traditional positioning methods use hard clustering patterns for region division, such as... Figure 1AAs shown, a "one-size-fits-all" hard clustering approach is adopted. This method is simple and direct, but it can easily lead to awkward classification for TP1 located at the inter-cluster boundary, and even cause TP1 located in cluster Ω to be classified incorrectly. s TP2, whose stability maintenance capabilities are relatively weak in the peripheral areas, was incorrectly assigned to the nearest neighbor cluster Ω. t middle.

[0108] This application employs a soft clustering model, which differs from the traditional hard clustering model in that it allows samples to belong to multiple prototypes simultaneously. Considering the misjudgment process of similarity between sample point clusters, a fuzzy space Ω is added to the transition region, based on the original "single" region, to weaken the absolute difference at the boundary. s &Ω t ,like Figure 1B As shown, the matching range is conditionally expanded to assign edge TPs to the fuzzy space, thus avoiding misclassification of TPs with insignificant differences in cluster similarity.

[0109] (1) Dual-attribute feature

[0110] Without loss of generality, the RP in the planar space is initially clustered using only its signal domain mean as a single scale, generating T clusters Ω1,Ω2,...,Ω T And record its cluster centers c1, c2, ..., c T Considering that cluster edge samples typically have a lower degree of belonging to the cluster, their regional discrimination is more susceptible to signal fluctuations. Therefore, an update index for the influence of signals on spatial discrimination is established to solve the problem of misjudgment of the similarity between sample points in the transition zone.

[0111] 1) Inter-class sparsity

[0112] The difference between a fingerprint point and fingerprints of different classes is measured by the feature distance between fingerprint point classes. The inter-class distance b of fingerprint points is defined as follows. (t) (s,i) represents the i-th fingerprint point in the s-th class. up to all n in class t t The average signal domain distance of each fingerprint point, i.e.

[0113]

[0114] Where s and t represent class labels, s,t∈{1,2,...,T}, s≠t, and They represent and The received signal sample value. (b) (t) (s,i) identifier Compared to Ω t The smaller the inter-class sparsity, the less obvious the differences in fingerprint information. Belonging to Ω t The higher the probability, the better.

[0115] 2) Intraclass tightness

[0116] Fingerprint information within the same class reflects the intrinsic similarity between samples. The intra-class feature distance of fingerprint points is typically used to explore the closeness between targets. Let w(s,i) represent the intra-class distance of the i-th fingerprint point in the s-th class. To the other n in this class s -1 is the average signal domain distance of fingerprint points, i.e.

[0117]

[0118] in, express The received signal sample value, and j≠i. At this time, with The intra-class distance maps to its intra-class density. The larger the value of w(s,i), the lower the intra-class density, and the further the fingerprint point is from the cluster center.

[0119] (2) Soft clustering generation

[0120] When the inter-class sparsity and intra-class compactness of a RP are low, it reflects that the RP's ability to maintain spatial signal stability is not outstanding. Inter-class distance b (t) The smaller (s,i) is, the larger the intra-class distance w(s,i) is, which is why it originally belonged to Ω. s of The more likely it is to be misjudged as Ω t To mitigate the risks, this application uses the dual distance of RP "inter-class-intra-class" to characterize the ambiguity of transition areas in a scene, incorporating RPs with indistinct fingerprint differences between inter-class and intra-class areas into the ambiguity region, thereby achieving cross-domain attribution.

[0121] Furthermore, to balance the discrimination accuracy and expansion cost of sub-regions, the ambiguity of the matching needs to be dynamically adjusted according to environmental characteristics. Therefore, a spatial ambiguity threshold f is introduced to measure... When there is a possibility that the region belongs to a fuzzy area, the absoluteness of the region can be flexibly adjusted to seek a fuzzy threshold with spatial uniqueness.

[0122] By utilizing the dynamic differences between signal and space, a given sub-region Ω is established. t The update model is shown in equation (3), which determines... Should it be added to subregion Ω? t Inside.

[0123]

[0124] It is easy to see that the larger the f value, the more RPs that satisfy spatial ambiguity, and the lower the probability of region misjudgment.

[0125] The dual-distance difference soft clustering algorithm uses a spatially overlapping classification structure to solve the target region attribution problem based on RSS fingerprints. On the basis of the initial fingerprint division constructed by the general hard clustering algorithm, it completes the dual-distance difference comparison of all RPs in the planar space, determines the fuzzy category, and thus realizes the transformation of the entire target environment from hard clustering to soft clustering.

[0126] For example, hard clustering algorithms will cluster RP1 to RP2. 10 These 10 reference points are assigned to region 1, and RP 11 ~RP 20 These 10 reference points are assigned to region 2, while RP 10 Located at the boundary between Zone 1 and Zone 2, when the user is online in RP 10 When locating a location in the vicinity, due to the significant signal fluctuations, it is easily classified into region 2. In this case, the location to be determined is only associated with the offline assigned RP within region 2. 11 ~RP 20 Matching these 10 points inevitably increases the positioning error. This invention, however, judges the signal stability of each reference point (RP) based on hard partitioning, and performs fuzzy processing on RPs with poor stability, allowing them to belong to multiple sub-regions simultaneously. For example, consider the RPs listed below. 10 By applying this invention, the location is simultaneously divided into region 1 and region 2. When positioning is performed near this location, even if region 2 is mistakenly identified, the point to be located can still be simultaneously aligned with RP. 10 ~RP 20 Comparing these 11 points helps to mitigate positioning errors caused by regional misjudgments to some extent.

[0127] Step 3: Constructing a dual-label fingerprint database;

[0128] The fuzzy partitioning and category reconstruction process in planar space ensures the robustness of region attribution for each RP during the offline fingerprint database construction. To improve the timeliness of online localization, the offline fingerprint database is constructed using class labels (or region labels) and intra-class point labels.

[0129] Clustering subclusters Ω1,Ω2,...,Ω based on RSS mean vector T As class labels; due to the fuzzy space setting, the same RP may belong to multiple classes simultaneously. Therefore, the updated set of RPs contained in each class is used as the in-class label, and the points belonging to Ω are assigned to Ω. t The inner point label of the class is denoted as Where G represents Ω t The updated number of intra-class point tags. Class tags, intra-class point tags, and their fingerprint information are jointly constructed into a dual-tag offline fingerprint database.

[0130] Step 4: Calculate the location to obtain the discrimination sub-region and estimated location of the point to be located.

[0131] (1) District affiliation

[0132] To avoid positioning errors caused by mismatches in the cluster edge regions, the planar spatial soft clustering algorithm introduces a local discrimination process that allows for boundary overlap. This overcomes the problem of poor edge judgment caused by RSS signal fluctuations. By softly partitioning the transition region, the interference from unstable inter-cluster boundary signals is weakened, improving the generalization ability of the positioning. To balance region discrimination accuracy and time complexity, when identifying region affiliation online, the signal domain is used as the scale, and the 1NN algorithm is employed to match the nearest sub-cluster Ω for TP. t As the home region, Ω t The RP set contained within As candidate fingerprint points.

[0133] (2) Point sorting

[0134] Considering that the closer the TP and RP are, the more similar signal strength values ​​will be received under more APs, converting the RSS difference between RP and TP into a relative ranking can reduce the impact of AP uncertainties on positioning. However, ranking errors caused by the dynamic characteristics of the signal are still difficult to avoid. Based on this, this invention no longer limits itself to measuring the similarity between the point to be located and the fingerprint point by the Euclidean distance between points. Instead, it combines the positioning information carried by the nearest neighbor space of the RP, weakens the influence of signal fluctuations, and performs a robust ranking of the RP based on homogeneous similarity, thereby achieving an effective judgment of multi-source balanced similarity with the TP.

[0135] 1) Homology similarity

[0136] Since the absolute difference in received signal strength between the target point TP and the reference point RP can reflect the similarity between the target point and the fingerprint point, within the same AP... m Below, for those already assigned to Ω t TP, calculate its relationship with Ω t middle The absolute difference in received signal strength is used to map the similarity of origins:

[0137]

[0138] in, The larger the value, the more likely TP is to be related to The higher the spatial proximity, the better. To avoid the occasional case where the denominator is 0, the additional term σ > 0 and σ → 0.

[0139] 2) RP homology sorting

[0140] The dynamic characteristics of a scene and its temporal differences lead to inconsistencies in the features of individual access points (APs) between offline and online phases, and can even result in a surge in absolute differences, causing misjudgments of the proximity between top-performing (TP) and bottom-performing (RP) connections. To ensure consistency in AP comparisons, the quantitative comparison of source similarity is transformed into a ranking comparison of source similarity to avoid significant differences in magnitude between different APs. The principle of RP source ranking is defined as follows:

[0141] For the same source AP m , TP and Ω t All RP pairs within AP m The similarity set is denoted as This is used to obtain the fingerprint points corresponding to the elements of the set in descending order. At the source AP m Relative sorting values

[0142] 3) Wild value correction

[0143] The interplay between signal fluctuations and the ranking results of other reference points under the same source AP can lead to anomalies in the ranking of some reference points (RPs). Given that RPs that are similar in the low-dimensional spatial domain usually have more similar RSS values ​​in the high-dimensional signal domain, neighboring RPs under the same source AP should have similar rankings. If the ranking of a certain RP differs significantly from that of its surrounding neighbors, the probability of that RP ranking being an outlier is higher.

[0144] Inspired by the method of filtering outliers using box plots, samples from the same source neighbor are sorted in ascending order and then subjected to quartile processing, denoted as Q. u and Q d The upper and lower quartiles are the interquartile range (IQR) = Q. u -Q d Based on Q d and Q u Scaling the IQR by 1.5 times to divide the effective interval, constructing the home region Ω. t Outlier correction model for internal RP homology sorting:

[0145]

[0146] In the formula, express The non-abnormal sorting (satisfying the valid interval) of neighboring points. If Located outside the boundary, without losing its generality, it is updated to be The average of the sorted values ​​is used to weaken the interference of outlier values ​​in the positioning; otherwise, Keep the original value.

[0147] 4) RP multi-source sorting balance

[0148] To systematically evaluate the ranking value of homology similarity, all AP pairs The ranking effect is considered in a multi-source equilibrium assessment, as... Final sorting:

[0149]

[0150] It should be noted that the non-integer sorting results obtained after normalization still follow the sorting rules.

[0151] (3) Point matching

[0152] Ideally, the K nearest neighbors (RPs) used for final location determination should surround the target point (TP). However, due to dynamic environmental interference and signal fluctuations, the K nearest neighbors selected by traditional KNN and WKNN matching algorithms cannot guarantee a clustered distribution based on the TP, and fixing the K value reduces the algorithm's adaptability to the environment. Therefore, this paper proposes a localization method that iteratively constrains the nearest neighbor set in both the signal and spatial domains. By introducing spatial density reachability criteria, a strongly correlated RP set is searched, and the K value is adaptively controlled, improving the accuracy of nearest neighbor RP selection while ensuring its relatively concentrated distribution.

[0153] 1) Search for strongly correlated RP sets

[0154] Obtaining the home region Ω using point sorting method t From all the sorted results of RP and TP, take the first I (I≤G) reference points to generate the initial nearest neighbor set RP'={RP'1,RP'2,...,RP' i ,...,RP' I Considering the uncertainty in the signal domain, some outliers will be introduced, such as... Figure 2A As shown, in order to overcome the adverse effects of outlier points on the localization results, based on the idea of ​​density clustering, the spatial domain distance of the initial screening nearest neighbor points is used to perform secondary constraints on the target nearest neighbor set.

[0155] To screen nearest neighbor RP' i Define RP' with the spatial coordinates as the center and ε as the radius. i The neighborhood is:

[0156] Nei ε (RP' i )={RP' j ∈RP'|dist(RP' i ,RP' j )≤ε} (7)

[0157] Among them, dist(RP') i ,RP' j ) represents two initially screened nearest neighbors RP' in set RP'. i and RP'j Spatial domain distance between them, Nei ε (RP' i ) contains RP' and RP' i All initial nearest neighbor points whose spatial distance is not greater than ε.

[0158] Set a threshold η in the domain to reflect RP' i The density of surrounding neighboring points, if |Nei ε (RP' i If |≥η, then RP' i This is considered a core point (CP). If a core point sequence exists... Then, according to equation (8), it is determined whether the preceding term of an adjacent core point in the sequence is within the neighborhood of the following term, and labeled with l. h Logo:

[0159]

[0160] The resulting H-1 labels collectively determine the core point sequence {CP1, CP2, ..., CP...} h ,...,CP H Whether a set of neighbors has the potential to form a target nearest neighbor set is defined as follows:

[0161]

[0162] If L = 1, then CP H If the distance to CP1 density is reachable, it indicates that the spatial domain of the core point sequence is relatively close, and the fingerprint information it carries has a high degree of similarity, making it possible to form a target nearest neighbor set; otherwise, it does not meet the conditions for forming a set.

[0163] Starting from any given CP, a cluster is formed from all CPs that are density-reachable from that CP; then, new starting points for clustering are found from the remaining CPs that are not yet clustered, thus completing the search for the candidate set of strongly correlated RPs, such as... Figure 2B As shown, points that cannot form clusters are considered weakly correlated reference points. Clearly, the neighborhood radius ε and the neighborhood threshold η jointly constrain the correlation of the RP candidate set.

[0164] 2) Target nearest neighbor matching

[0165] To further ensure the reliability of adaptive matching of the target nearest neighbor set, the target nearest neighbor set is constrained three times by using the number of CPs in the strongly correlated RP set and their average signal domain distance to TP, thus selecting the target set with the highest probability. If a strongly correlated RP set has the smallest average signal domain distance to TP and contains the most CPs, it is retained as the target nearest neighbor set. Where K (2≤K≤I) represents the number of RPs in the target nearest neighbor set, such as Figure 2C As shown; otherwise, it indicates that the strongly correlated RP set is insufficient to support the determination of the point to be located, and the initial screening nearest neighbor set is output for the final location estimation.

[0166] It should be noted that there is a special case where the initial screening nearest neighbor set of a certain TP is entirely judged as a weakly correlated reference point. In this case, the initial screening nearest neighbor set is also used as the target nearest neighbor set for the location estimation of the TP.

[0167] (4) Location calculation

[0168] The multi-source ranking equalization result of RP reflects the level of similarity between the received signals of RP and TP. Therefore, it is used as the basis for configuring TP positioning weights, and is characterized as follows:

[0169]

[0170] With a target nearest neighbor set that has strong convergence As a valid reference set for TP location estimation, the planar spatial coordinates (x', y') of the point TP to be located are estimated:

[0171]

[0172] Example 3

[0173] This embodiment provides an indoor positioning method for overcoming signal interference in home healthcare, applied to positioning within a relatively large planar space in a home care setting.

[0174] To evaluate the method's performance, a high-traffic polygonal corridor (in practical applications, this could also be a senior care facility) was used as the real-world test site. Reference points were deployed in a uniform grid, with adjacent reference points (RPs) spaced 1m apart, for a total of N = 270 RPs. Each RP received signals from M = 86 access points (APs), with undetected AP signal strength values ​​replaced by -100dB. Fingerprint data was collected Q = 60 times at a 2.3s sampling interval. 152 online test points (TPs) were selected along the central area of ​​the experimental corridor.

[0175] To avoid the impact of device differences on algorithm performance evaluation, the same mobile terminal was used as the data acquisition tool for both offline and online processes.

[0176] To evaluate the impact of configuring the optimal fuzzy threshold f, experiments were conducted comparing the changes in region discrimination accuracy and global RP growth rate under different f values, such as... Figure 4 As shown, when f=11, the region discrimination accuracy reaches 99.68%, while the global RP growth rate is only 10.37%. Considering the feasible range and computational burden of searching strongly correlated RP sets, without loss of generality, 20% of the total number of RPs in the sub-region is used as the number of initial screening nearest neighbors.

[0177] Furthermore, to optimize the system's positioning performance, the neighborhood radius ε and neighborhood threshold η are optimized to obtain the impact of multiple configuration schemes of ε∈{1,1.5,2,...,4.5,5} and η∈{1,2,...,10} on positioning performance in each sub-region. Figures 5A-5C The localization performance of subregions Ω2, Ω5, and Ω7 under different combinations of neighborhood parameters is presented. Experiments show that different combinations of neighborhood parameters (ε, η) result in different localization performances for the subregions.

[0178] Considering the time-sensitivity of clustering algorithms, this invention selects the conventional K-means algorithm as the initial region partitioning algorithm, forming 7 independent regions and 6 fuzzy regions, as follows: Figure 6 As shown, the soft labeling property of DLFS is reflected in the overlapping division of 28 reference points (RPs) with small differences in bi-distance. Experiments show that RPs with fuzzy division properties are mostly located at the boundaries of sub-regions. The assignment results of these RPs to bi-regions further confirm the insufficient robustness of the transition zone signal. Due to spatial structure changes and signal overlap, the boundary area is more prone to regional misjudgment. Obviously, the setting of fuzzy space can weaken the influence of regional signal instability and control regional positioning error. Compared with global greedy matching, the soft clustering method proposed in this application reduces the number of reference points N=270 in the entire target area to 24-53 in the local area, which largely ensures the real-time positioning. In addition, the overlapping division of 28 reference points only adds 1-7 limited reference points to each sub-region, which is relatively controllable compared to the local matching workload added by the strict hard division algorithm.

[0179] To verify the overall localization performance of the proposed DLFS localization method, it was compared with six similar algorithms. Among the six similar algorithms, the traditional WKNN algorithm, EWKNN, and AAS are three global matching algorithms:

[0180] For the WKNN method, please refer to YANG H, ZHANG Y, HUANG Y, et al. WKNN indoor location algorithm based on zone partition by spatial features and restriction offormer location[J]. Pervasive and Mobile Computing, 2019, 60(10):1-14.

[0181] For the EWKNN method, please refer to Wang Peizhong, Zheng Nanshan, Zhang Yanzhe. Indoor positioning algorithm based on dynamic K value and AP MAC address filtering [J]. Computer Science, 2016, 43(1):163-165.

[0182] For the AAS method, please refer to Tao Y, Zhao L. Fingerprint Localization with Adaptive AreaSearch[J]. IEEE Communications Letters, 2020, 24(7):1446-1450.);

[0183] In addition, three different offline region segmentation algorithms, namely APC, FCM and K-means, are added and combined with the online method of this invention to achieve indoor positioning.

[0184] For the APC method, please refer to Tian Z, Tang X, Zhou M, et al. Fingerprint indoorpositioning algorithm based on affinity propagation clustering[J]. EURASIPJournal on Wireless Communications and Networking, 2013, 2013(1):1-8.

[0185] For the FCM method, please refer to Zhou H, Van N N. Indoor fingerprint localization based on fuzzy c-means clustering[C] / / In Proceedings of the 2014Sixth International Conference on Measuring Technology and Mechatronics Automation (ICMTMA), Zhangjiajie, China, 10–11January 2014; 337–340.

[0186] For the K-means method, please refer to Altintas B, Serif T. Improving RSS-Based indoorpositioning algorithm via K-Means clustering[C] / / In Proceedings of the European Wireless, Vienna, Austria, 27–29 April 2011; 681–685.).

[0187] Figure 7 shows the cumulative probability distribution of positioning errors for a total of seven algorithms, including global matching and local matching. When estimating the TP position, within the same accuracy range, the positioning method DLFS proposed in this application outperforms all other comparative algorithms. In comparison with global matching, the DLFS algorithm achieves an estimation accuracy within 1.4m in 73.78% of cases, representing a 21.35% improvement in estimation accuracy compared to the WKNN algorithm at that time. Figure 7A As shown. By Figure 7B It can be seen that as the positioning error range expands, the cumulative error probability of the four local matching methods tends to increase in a consistent manner. When the positioning error range expands to 0.4m, the DLFS algorithm shows its advantage, and when it expands to 2.4m, the cumulative error probability reaches 90.00%. Experiments demonstrate that DLFS has significant advantages in solving positioning problems caused by environmental and signal dynamic characteristics. This is mainly due to the sensitivity of the fuzzy clustering mechanism in handling boundary signal jump problems, and the efficient measurement of the similarity between fingerprint points and the point to be located by the online positioning method.

[0188] This application method utilizes existing resources, leveraging low-cost, low-power wireless access devices to meet users' location needs in home-based medical monitoring. After fingerprint collection and database creation for the target location scenario, the area is divided. When a user requests location services, the system first determines the user's likely location, such as a room in a home or a corridor in a hospital. After determining the likely location, the system estimates the user's specific position within that area. This method can be applied to various indoor scenarios, including hospitals, nursing homes, communities, and homes. For example, if children living away from home need to monitor elderly individuals living alone, accurate location tracking can determine the elderly person's location and track their movement. If the elderly person remains stationary at home, excluding activities like sleeping, watching TV, or using the toilet, they might fall and be unable to help themselves. This can trigger the home healthcare alarm system and allow for targeted medical assistance based on the elderly person's specific location. Similarly, in hospitals, if elderly individuals or children wander out of their rooms due to caregivers' negligence, accurate location tracking becomes crucial. An efficient indoor positioning system can bring greater convenience and prevent numerous problems.

[0189] Some steps in the embodiments of the present invention can be implemented using software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk.

[0190] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An indoor positioning method for overcoming signal interference in home-based medical care, characterized in that, The method includes an offline phase and an online phase; The fingerprint database is constructed offline, and initial clustering of reference points in the database is performed using average signal strength as a single scale, generating... Each cluster is defined and its corresponding cluster center is recorded. Reference points located at the edge of the cluster are then subjected to secondary clustering using the "between-intra-cluster" distance. In the online phase, the RSS vector received in real time by the point to be located (TP) is matched with the samples in the fingerprint database constructed in the offline phase to estimate the spatial location of the target. During the matching process, the similarity between the point to be located and the fingerprint points is measured by the Euclidean distance between the points. At the same time, the positioning information carried by the nearest neighbor space of the RP is combined to weaken the influence of signal fluctuations. The RP is robustly ranked according to the similarity of the same source to achieve accurate estimation of the spatial location of the target.

2. The method according to claim 1, characterized in that, The method includes: Step S1: Construct the system model; Offline phase: Assume that there are N reference points RP and M access points AP deployed in the planar space where the point to be located is located; The physical location is denoted as ,in, and in various The above samples Q signal strength values ​​from M non-line-of-sight access points (APs), denoted as . , , ; Then in The location received information from the source. The average signal strength is ,by The average signal strength characterizes the positional features of the location point and is represented as a vector. And record fingerprint information The fingerprint database contains fingerprint information for all reference points (RPs). Online phase: Located in The signal strength sample values ​​received by the target location TP from M access points AP are denoted as follows: ; Step S2, spatial fuzzy partitioning; Initial clustering is performed using the average signal intensity at each reference point as a single scale to generate... Cluster And record its cluster center Reference points belonging to the same cluster are called a class; the planar space is divided according to the class to which they belong. Sub-regions ; Calculate the inter-class distances of each reference point and intra-class distance ; For the first The first in the class fingerprint points To the All in the class Average signal domain distance of each fingerprint point; and Indicates class tags, ; Represented as the first The first in the class fingerprint points To other in this class Average signal domain distance of each fingerprint point; Set spatial blur threshold Based on the inter-class distance of each reference point and intra-class distance The absolute value of the difference and the spatial fuzziness threshold Relationship Determination of Reference Point Should it be added to the sub-region? Internally, implement category updates; Step S3: Construct a dual-label fingerprint database; The sub-clusters generated during the initial clustering partitioning As the class label for the target region, the updated RP set contained in each class is used as the in-class point label, and the points belonging to each class are assigned to the target region. The inner point label of the class is denoted as , ,in express The updated number of in-class point labels; A dual-label offline fingerprint database is constructed by combining class labels, intra-class point labels, and their fingerprint information. Step S4: Position calculation, to obtain the discrimination sub-region and estimated position of the point TP to be located; The signal strength values ​​from each AP are collected at the point to be located (TP), and the nearest sub-cluster is matched to TP based on the collected signal strength values. As the area of ​​belonging, The RP set contained within As candidate fingerprint points; For those already assigned TP, calculate its relationship with middle The absolute difference in received signal strength is used to map the similarity of origin. ; For the same information source , TP and All RP pairs The similarity set is denoted as This allows us to obtain the fingerprint points corresponding to the elements in the set in descending order. In the source Relative sorting values ; To systematically evaluate the ranking value of origin similarity, all access points are paired with reference points. The ranking effect is considered in multi-source equilibrium. As The final sorting; Introducing spatial density reachability discriminant, searching for strongly correlated RP sets, and controlling... Value adaptation determines the effective reference set of the point TP to be located. Then, estimate the planar spatial coordinates of the point TP to be located. : (11) in, , Indicates the effective nearest neighbor. The sorting value.

3. The method according to claim 2, characterized in that, When performing spatial fuzzy partitioning in step S2, the inter-class distance... The calculation formula is: (1) in, and They represent and The received signal sample value; Intra-class distance The calculation formula is: (2) in, express The received signal sample value, and .

4. The method according to claim 3, characterized in that, The inter-class distance based on each reference point and intra-class distance The absolute value of the difference and the spatial fuzziness threshold Relationship Determination of Reference Point Should it be added to the sub-region? Internally, category updates are implemented, including: (3)。 5. The method according to claim 4, characterized in that, The homology similarity in step S4 The calculation formula is: (4) Among them, additional items and .

6. The method according to claim 5, characterized in that, In step S4, the fingerprint points corresponding to the elements of the set are obtained in descending order. In the source Relative sorting values Then, it also includes: sorting the samples from the same source neighbor in ascending order and then performing quartile processing, denoted as... and The upper and lower quartiles, interquartile range ,based on and Scaling Divide the effective interval by 1.5 times the value to construct the belonging region. Outlier correction model for internal RP homology sorting: (5) In the formula, express Non-abnormal sorted neighbor points; like Located outside the boundary, update it to The average of the sorted values; otherwise, Keep the original value.

7. The method according to claim 6, characterized in that, In step S4, spatial density reachability discrimination is introduced to search for strongly correlated RP sets and control... Value adaptation determines the effective reference set of the point TP to be located. ,include: 1) Search for strongly correlated RP sets Get Home Zone The sorting results of all RP and TP are taken from the top. I Using reference points, generate an initial screening nearest neighbor set. ; The spatial domain distance of the initial nearest neighbor points is used to apply secondary constraints to the target nearest neighbor set; To screen nearest neighbors Centered on the spatial coordinates, Define the radius. The neighborhood is: (7) in, Represents a set Two primary screening nearest neighbor points and Spatial domain distance between them Includes Zhongyu Spatial domain distance not greater than All of the initial screening nearest neighbors; Set domain threshold To reflect The density of surrounding neighboring points, if Then Consider it as a core point, CP; if a core point sequence exists. Then, according to equation (8), it is determined whether the preceding term of an adjacent core point in the sequence is within the neighborhood of the following term, and the label is used to determine whether the preceding term is within the neighborhood of the following term. Logo: (8) Formed The labels together determine the core point sequence. Whether it has the potential to form a target nearest neighbor set, defined as: (9) like ,but distance If the density is achievable, it indicates that the spatial domains of the core point sequences are relatively close, and the fingerprint information they carry has a high degree of similarity, making it possible to form a target nearest neighbor set; otherwise, it does not meet the conditions for forming a set. Starting from any core point CP, a cluster is formed from all CPs that the density of that CP can reach; then, new starting points that can be clustered are found from the remaining CPs that have not yet been clustered, thus completing the search for the candidate set of strongly correlated RPs. Points that cannot be clustered are regarded as weakly correlated reference points. 2) Target nearest neighbor matching Using the number of target CPs in a strongly correlated RP set, supplemented by their average signal domain distance to the target TP, a triple constraint is applied to the target nearest neighbor set to select the set of targets with the highest probability: If a strongly correlated RP set has the smallest average signal domain distance to TP and contains the most CPs, then it is retained as the target nearest neighbor set. ,in This represents the number of nearest neighbors (RP) in the target neighborhood set. Otherwise, the initial nearest neighbor set is output for final location estimation.

8. The method according to claim 7, characterized in that, If the initial screening nearest neighbor set of a certain TP is entirely judged as a weakly correlated reference point, then the initial screening nearest neighbor set is used as the target nearest neighbor set for the location estimation of that TP.

9. The application of the indoor positioning method according to any one of claims 1-8 in rehabilitation and elderly care.

Images